Support RealSR NCNN Vulkan for Face reconstruction Human Eyes from DVD Upscaling


There is the AI Algorithm Realsr Ncnn Vulkan, which scales each individual frame. The image quality of the video is heavy. The disadvantage is economy. Even with an Asus Turbo 1080 Ti you only get 0.02fps. I tested it myself with my workstation (Intel Core i9-9900K, Asus Turbo 1080 Ti) and Video2x (OpenSource Machine Learning Video UPSCALING)

What do I want to go ???

I am currently trying to restore DVD’s in BluRay quality. I use Dione TV4. The problem that results from this is the following. The eyes of some people in the background or poorly exposed scenes are strongly pixelated (see file attachment)

I will soon be at the end with my Latin. I could also disassemble the source video into individual images, then send them through an AI image software and then render them again. But honestly, who does that, nobody.

I’m not going to vote for your suggestion, but I understand the want to not need more programs to be able to reach your goal of DVD to Blu-ray quality.
Your examples look like you have interlaced sources. Most people on these forums agree that non-AI programs do a better job at de-interlacing. (Like ffmpeg or staxrip.)

What I have found through extensive trials of different approaches, is the two best models for DVD to Blu-ray quality, are Gaia HQ and Proteus—But only if the source has low to no noise.
To clean the source, I use Artemis Medium Quality at 100% scaling. It works really well, but I have not tested it on sources with lots of compression artifacts.
After that, I run Gaia or Proteus on it doing the upscaling.

Anyway, those are the general steps. It sure would be nice if there were a model that could produce even better results than that in one run, and without help from other programs, but VEAI is not there yet.